基于改进烟花算法优化极限学习机的过热汽温特性建模

Modeling of Superheated Steam Temperature Characteristics based on Extreme Learning Machine Optimized by Improved Fireworks Algorithm

为建立精确的过热汽温对象模型,以实现过热汽温的智能预测优化控制。借助某600 MW亚临界机组DCS历史运行数据,采用具有外部时延的极限学习机(extreme learning machine,ELM)建立了该锅炉过热汽温特性的预测模型,并采用引入自适应调整惯性权重的烟花算法(Improved Fireworks algorithm,IFWA)对模型参数进行优化,将IFWA-ELM模型与标准ELM模型的预测结果进行对比。结果表明:针对某600 MW亚临界机组,改进的烟花算法鲁棒性强、收敛结果更准确,优化后一、二级过热汽温特性预测模型的测试集平均相对误差分别为0.1919%和0.097%,具有更好的预测精度与泛化能力。

In order to establish a more accurate superheated steam temperature(SST)characteristic model and realize intelligent predictive optimal control of SST,the prediction model of SST was established with external time-delay extreme learning machine(ELM)method using the historical operation data acquired from DCS of a 600 MW subcritical coal-fired power unit.The improved fireworks algorithm(IFWA)with adaptive inertia weight was adopted to optimize the ELM neural network parameters.By comparing the model prediction results of IFWA-ELM model with those from the traditional ELM model,it is shown that the improved fireworks algorithm is more robust and the convergence result is more accurate for the 600 MW subcritical unit,and the MRE of the 1st and 2nd stage SST characteristic prediction models optimized by IFWA is 0.1919%and 0.097%respectively,indicating better prediction accuracy and generalization ability.